i think everything everywhere in the internet will be put to training AI at this point. Lemmy and other FOSS will be used too, but at least our data is public and accessible to everyone equally (including to some FOSS AI that i hope emerges), not a private property of someone.
That is the true beauty of FOSS technology. Even if it fractures into regional forks, Linux code is open and free (as in freedom), so each fork can just copy-paste and compile the changes made in others if they advance the tech forward, no direct cooperation is actually needed (if everyone keeps publishing its works).